Open-source entity coreference resolution systems

English[]

System	Syntax feature? ^{[note 1]}	Semantic role feature?	Semantic type feature? ^{[note 2]}	Word-window feature?	Mention-pair?	Entity-mention?	Mention-ranking?	Cluster-ranking?	Cluster-pair?	Rule-based?	Base ML model	Integer linear programming?	Reference	Notes
cort	Yes ^{[note 3]}	No	Yes ^{[note 4]}		Yes	Yes	Yes	No	No	No	perceptron	No	Martschat and Strube (2015)^[1]
nn_coref	Yes ^{[note 5]}	No	Yes ^{[note 6]}		No	Yes	No	No	No	No	neural net (RNN for encoding clusters)	No	Wiseman et al. (2016)^[2]
huggingface's neural coref	Yes	No	Yes ^{[note 7]}	Yes ^{[note 8]}	No	No	Yes	No	No	No	neural net	No	medium post	impl. of Clark and Manning (2016)^[3]
deep-coref(CoreNLP)	Yes	No	Yes ^{[note 7]}	Yes ^{[note 8]}	Yes	No	Yes	Yes	No	No	neural net	No	Clark and Manning (2016)^[4]
hcoref (Hybrid Coref)	Yes ^{[note 9]}	No	Yes ^{[note 10]}	No	No	Yes?	No	No	Yes?	Yes ^{[note 11]}	random forest ^{[note 11]}	No	Lee et al. (2017a)^[5]
dcoref(Stanford sieve)	Yes	No	Yes ^{[note 12]}	No	No	No	No	No	Yes	Yes	None	No	Lee et al. (2013)^[6]	part of Stanford CoreNLP
Berkeley CR	No	No	Yes	No	No	Yes ^{[note 13]}	Yes ^{[note 14]}	No	No	No	log-linear	No	Durrett and Klein (2013)^[7]
Illinois CR										No		Yes
xrenner										Yes	None	No	Zeldes and Zhang (2016)^[8]
e2e-coref	No	No	No	No	No	No	Yes	No	No	No	feed-forward NN + LSTM + attention	No	Lee et al. (2017b)^[9]
allennlp	No	No	No	No	No	No	Yes	No	No	No	feed-forward NN + LSTM + attention	No		a reimplementation of e2e-coref with some changes

German[]

System

Syntax feature?

^{[note 15]}

Semantic role feature?

Semantic type feature?

^{[note 16]}

Word-window feature?

Mention-pair?

Entity-mention?

Mention-ranking?

Cluster-ranking?

Cluster-pair?

Rule-based?

Base ML model

Integer linear programming?

Reference

Notes

CorZu

University of Zurich

HotCoref

University of Stuttgart

Notes[]

↑ Meaning the syntactic relation between mentions or between mention and surrounding words. Head word features (which might come from a parser) is not considered a syntactic feature.
↑ Different from semantic role features, this includes features about mentions alone: semantic type (person/object/number), NER type (person/location/organization), or other taxonomies.
↑ deprel: dependency relation of a mention to its governor
↑ sem_class: one of 'PERSON', 'OBJECT', 'NUMERIC' and 'UNKNOWN' and head_ner: named entity tag of the mention's head word
↑ From syntactic ancestry features in BASIC+ (Wiseman et al. 2015)
↑ From entity type features in BASIC+ (Wiseman et al. 2015)
↑ ^7.0 ^7.1 In Clark and Manning (2016): "The type of the mention (pronoun, nominal, proper, or list)"
↑ ^8.0 ^8.1 From Clark and Manning (2016): "first word, last word, two preceding words, and two following words of the mention. Averaged word embed- dings of the five preceding words, five following words, all words in the mention, all words in the mention’s sentence, and all words in the mention’s document."
↑ Feature: "The path in the parse tree from the root to the (antecedent/anaphor)"
↑ Feature: "named entity type attributes of (antecedent/anaphor)"
↑ ^11.0 ^11.1 They combine rule-based and statistical classifiers.
↑ "NER label – fromthe Stanford NER"
↑ TRANSITIVE model: "each mention to maintain its own distributions over values for a number of proper- ties; these properties could include gender, named- entity type, or semantic class. Then, we will require each anaphoric mention to agree with its antecedent on the value of each of these properties"
↑ BASIC model: "This approach is similar to the mention- ranking model of Rahman and Ng (2009)."
↑ Meaning the syntactic relation between mentions or between mention and surrounding words. Head word features (which might come from a parser) is not considered a syntactic feature.
↑ Different from semantic role features, this includes features about mentions alone: semantic type (person/object/number), NER type (person/location/organization), or other taxonomies.

References[]

↑ Sebastian Martschat and Michael Strube. 2015. Latent structures for coreference resolution. TACL, 3:405– 418.
↑ Wiseman, Sam, Alexander M. Rush, and Stuart M. Shieber. "Learning Global Features for Coreference Resolution." arXiv preprint arXiv:1604.03035(2016).
↑ Clark, K., & Manning, C. D. (2016). Deep Reinforcement Learning for Mention-Ranking Coreference Models. Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP-16), 2256–2262.
↑ Clark, K., & Manning, C. D. (2016a). Improving Coreference Resolution by Learning Entity-Level Distributed Representations. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, 643–653. http://doi.org/10.18653/v1/P16-1061
↑ LEE, HEEYOUNG, MIHAI SURDEANU, and DAN JURAFSKY. "A scaffolding approach to coreference resolution integrating statistical and rule-based models." Natural Language Engineering (2017a): 1-30.
↑ Heeyoung Lee, Angel Chang, Yves Peirsman, Nathanael Chambers, Mihai Surdeanu and Dan Jurafsky. Deterministic coreference resolution based on entity-centric, precision-ranked rules. Computational Linguistics 39(4), 2013.
↑ Durrett, G., & Klein, D. (2013). Easy victories and uphill battles in coreference resolution. EMNLP ’13, (October), 1971–1982.
↑ Zeldes, A., & Zhang, S. (2016). When Annotation Schemes Change Rules Help: A Configurable Approach to Coreference beyond OntoNotes. Workshop on Coreference Resolution Beyond OntoNotes at NAACL, (Corbon), 92–101.
↑ Lee, K., He, L., Lewis, M., & Zettlemoyer, L. S. (2017b). End-to-end Neural Coreference Resolution. In EMNLP.

[1] Meaning the syntactic relation between mentions or between mention and surrounding words. Head word features (which might come from a parser) is not considered a syntactic feature.

[2] Different from semantic role features, this includes features about mentions alone: semantic type (person/object/number), NER type (person/location/organization), or other taxonomies.

[3] rel: dependency relation of a mention to its governor

[4] sem_class: one of 'PERSON', 'OBJECT', 'NUMERIC' and 'UNKNOWN' and head_ner: named entity tag of the mention's head word

[6] From syntactic ancestry features in BASIC+ (Wiseman et al. 2015)

[7] From entity type features in BASIC+ (Wiseman et al. 2015)

[:0-9] 7.0 ^7.1 In Clark and Manning (2016): "The type of the mention (pronoun, nominal, proper, or list)"

[:1-10] 8.0 ^8.1 From Clark and Manning (2016): "first word, last word, two preceding words, and two following words of the mention. Averaged word embed- dings of the five preceding words, five following words, all words in the mention, all words in the mention’s sentence, and all words in the mention’s document."

[13] Feature: "The path in the parse tree from the root to the (antecedent/anaphor)"

[14] Feature: "named entity type attributes of (antecedent/anaphor)"

[:2-15] 11.0 ^11.1 They combine rule-based and statistical classifiers.

[17] "NER label – fromthe Stanford NER"

[19] TRANSITIVE model: "each mention to maintain its own distributions over values for a number of proper- ties; these properties could include gender, named- entity type, or semantic class. Then, we will require each anaphoric mention to agree with its antecedent on the value of each of these properties"

[20] BASIC model: "This approach is similar to the mention- ranking model of Rahman and Ng (2009)."

[24] Meaning the syntactic relation between mentions or between mention and surrounding words. Head word features (which might come from a parser) is not considered a syntactic feature.

[25] Different from semantic role features, this includes features about mentions alone: semantic type (person/object/number), NER type (person/location/organization), or other taxonomies.

[:4-5] Sebastian Martschat and Michael Strube. 2015. Latent structures for coreference resolution. TACL, 3:405– 418.

[:2-8] Wiseman, Sam, Alexander M. Rush, and Stuart M. Shieber. "Learning Global Features for Coreference Resolution." arXiv preprint arXiv:1604.03035(2016).

[11] Clark, K., & Manning, C. D. (2016). Deep Reinforcement Learning for Mention-Ranking Coreference Models. Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (EMNLP-16), 2256–2262.

[:3-12] Clark, K., & Manning, C. D. (2016a). Improving Coreference Resolution by Learning Entity-Level Distributed Representations. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, 643–653. http://doi.org/10.18653/v1/P16-1061

[16] LEE, HEEYOUNG, MIHAI SURDEANU, and DAN JURAFSKY. "A scaffolding approach to coreference resolution integrating statistical and rule-based models." Natural Language Engineering (2017a): 1-30.

[18] Heeyoung Lee, Angel Chang, Yves Peirsman, Nathanael Chambers, Mihai Surdeanu and Dan Jurafsky. Deterministic coreference resolution based on entity-centric, precision-ranked rules. Computational Linguistics 39(4), 2013.

[21] Durrett, G., & Klein, D. (2013). Easy victories and uphill battles in coreference resolution. EMNLP ’13, (October), 1971–1982.

[22] Zeldes, A., & Zhang, S. (2016). When Annotation Schemes Change Rules Help: A Configurable Approach to Coreference beyond OntoNotes. Workshop on Coreference Resolution Beyond OntoNotes at NAACL, (Corbon), 92–101.

[23] Lee, K., He, L., Lewis, M., & Zettlemoyer, L. S. (2017b). End-to-end Neural Coreference Resolution. In EMNLP.

[note 1]

[note 2]

[note 3]

[note 4]

[1]

[note 5]

[note 6]

[2]

[note 7]

[note 8]

[3]

[4]

[note 9]

[note 10]

[note 11]

[5]

[note 12]

[6]

[note 13]

[note 14]

[7]

[8]

[9]

[note 15]

[note 16]

Open-source entity coreference resolution systems

Contents

English[]

German[]

Notes[]

References[]

Fan Feed